Please enable JavaScript.
Coggle requires JavaScript to display documents.
Ischemic Heart Disease, Note: They could have improved computation time if…
Ischemic Heart Disease
Leading cause of death
Deficiency of blood supply to ischemia
Early diagnosis could increase survivability
Electrocardiography
Used to monitor IHD
Had several advancements
Enhanced diagnostic performance for myocardial perfusion scintigraphy
for cardiac catheterization
Challenges
Invasive
Time-consuming
Requires technical experts
Use of magnetocardiograms
Contact-free
Non-invasive
Useful for monitoring magnetic fields emitted by cardiac tissues
Materials and Methods
MCG from 125 individuals
55 showed myocardial ischemia
70 were healthy
Acquired through 36 locations
4 measurements above the torso
Cardiac magnetic field were recorded for each segment
90s
Use of nine sensors
Sampling rate of 1000Hz
Post-processing
Digital low pass filter
Inputs to the model is the j-Point and T-peak interval of cardiac cycles
Network Architecture
Backpropagation Neural Network
Consists of Inputs, hidden layers, and output layers
Backpropagation is employed to increase performance
Dataset was normalized to a range of 0 to 1
Improve computation time
Removed redundant descriptors
Reduced to 2-78 from 1152
Optimal number is determined through trial and error
Used UFS software
Optimal number of hidden layers and nodes
Values with the lowest average root mean square error
Direct Kernel Self-Organizing Map
Used StripMiner software
Non-linear properties are stored in the hidden kernels
Utilized a U-Matrix
Results and Discussion
BNN optimization
12 descriptors yielded minimal predictive error
2 nodes gave the lowest RMSE
Converged in 350 epochs
Accuracy
DK-SOM = 80.4%
BNN = 74.5%
Specificity
DK-SOM = 72.7%
BNN = 54.5%
Sensitivity
DK-SOM = 86.2%
BNN = 89.7%
Note: They could have improved computation time if they trained using a GPU
Tantimongcolwat, T., Naenna, T., Isarankura-Na-Ayudhya, C.,
Embrechts, M., and Prachayasittikul, V. Identification of ischemic
heart disease via machine learning analysis on magnetocardiograms. Computers in biology and medicine 38 (08 2008), 817–25.